RDS Multi-AZ versus Read Replicas

WHAT THIS TESTS This probes whether you can tell availability apart from scalability, a distinction many engineers blur. Both features replicate data, but they exist for different reasons.

A GOOD ANSWER COVERS Multi-AZ maintains a synchronous standby instance in a different availability zone. The standby receives every write synchronously, serves no traffic, and exists purely for failover. If the primary fails, RDS flips the DNS endpoint to the standby automatically, usually within a minute or two. It improves durability and uptime, not performance. A read replica is an asynchronous copy you can query for read-only workloads. It offloads reporting, analytics, and heavy read traffic from the primary, and you can have several across regions. Because replication is asynchronous it can lag, so it is eventually consistent, and promoting it to a writable primary is a manual operation, not automatic failover. The two are orthogonal: Multi-AZ answers what happens when the primary fails, while read replicas answer how you scale read throughput, and modern RDS even lets a Multi-AZ cluster deployment expose readable standbys, blurring the line but not changing the core purposes.

COMMON WRONG ANSWERS Saying the Multi-AZ standby can serve reads to spread load; it cannot, it is passive. Claiming a read replica provides automatic failover; promotion is manual and the replica may be behind. Assuming replicas are always consistent with the primary.

LIKELY FOLLOW-UPS What is replication lag and how do you measure it? How do you route read traffic to replicas? What happens to in-flight transactions during a Multi-AZ failover? How does cross-region replication change cost and latency?

ONE CONCRETE EXAMPLE A payments service uses Multi-AZ so a zone outage causes a brief automatic failover with no data loss. As the analytics team starts running heavy dashboard queries, you add two read replicas and point all reporting and read-heavy endpoints at them, keeping the writable primary free for transactions. You now run both: Multi-AZ for resilience, replicas for read scale.